CD73 is compartmentalized in photoreceptor layer of the mouse retina, while CD39 is highly expressed in the eye vasculature, retinal microglia and cornea.
The first part of this study was designed to assess the tissue-specific distribution of nucleotide-inactivating enzymes in the naïve mouse eye. The use of sections of whole eyeball dissected from C57BL/6N mice and embedded in low melting point agarose (LMA) and their sequential incubation with antibodies against CD73, CD39, and NTPDase2 in combination with a wide range of molecular markers allowed us to characterize the phenotypic identity and spatial localization of key ecto-nucleotidases in a relatively thick (~ 100 µm) tissue volume. Staining of the eye with rabbit anti-rat CD73 (Fig. 1A-B) and rat anti-mouse CD73 (Supplementary Fig. 1A) antibodies recognizing different epitopes of CD73 revealed a similar pattern of selective compartmentalization of CD73 in the photoreceptor layer. The highest CD73 immunoreactivity was associated with the outer segments (OS) of photoreceptor cells, where it is co-localized with a light-sensitive receptor protein rhodopsin (a marker of rod cells) and blue-sensitive S-opsin (cone marker) (Fig. 1B and Supplementary Fig. 1A). Unlike CD73, CD39 is more broadly distributed in different ocular structures. High levels of CD39 were found in the retinal vasculature, including the central retinal artery and vein, which enter the optic nerve head and further bifurcate into smaller arterioles, venules and capillaries extensively branching throughout the inner (superficial) plexus and deeper capillary plexus, as well as in the choroid layer (choriocapillaris) and extraocular blood vessels (Fig. 1A and Supplementary Fig. 1B-C). Co-staining of the eyes with anti-CD39 antibody and different vascular markers demonstrated the presence of CD39 on all components of the vessel wall, including CD31+/IB4+ endothelial cells which share their basement membranes with adjacent NG2+/Phalloidin+ pericytes, and also contractile SMA-α+/Phalloidin+ smooth muscle cells (SMC) wrapped in a circumferential pattern around larger arterioles (Fig. 1C and Supplementary Fig. 2A-C). Interestingly, the close-up view of the images validated recent data on the presence of nanotube-like processes that connect two bona fide pericytes on separate capillary systems and regulate neurovascular coupling in the mouse retina , and further extend these observations by showing that these fine structures do not express CD39 and as a consequence, are unable to metabolize ATP (Fig. 1C, inset). The latter observation on unequal distribution of CD39 in retinal pericytes may be relevant in light of earlier findings showing the ability of extracellular ATP to induce pericyte contraction and capillary lumen narrowing only in a certain portion of retinal microvessels. . Significant CD39 immunoreactivity was also detected in the rhodopsin+ OS of photoreceptors (Fig. 1B), as well as Iba1+/P2Y12R+ microglial cells, which mainly reside in two synaptic compartments of the neural parenchyma: the outer plexiform layer (OPL) and the inner plexiform layer (IPL) (Fig. 1D and Supplementary Fig. 2A-B), and in the optic nerve head (Supplementary Fig. 1C). CD39 is also expressed, albeit faintly compared to blood vessels, photoreceptors, and microglial cells, on NeuN+ neuronal cell bodies located in the ganglion cell layer (GCL) (Fig. 1D), but not on vimentin+ intermediate filaments of Müller glia (Fig. 1A,D and Supplementary Fig. 1C) and GFAP+ positive astrocytes selectively compartmentalized in the innermost nerve fiber layer (NFL) (Supplementary Fig. 2B). CD39 is also co-localized with another member of the family, NTPDase2, on tubulin-βIII+ neuronal processes lining the innermost margin of the retina and cornea, as well as corneal IB4+/Phalloidin+ epithelial cells, and stromal keratocytes (Supplementary Fig. 2D).
The advantage of our workflow is that it provides additional information on high-resolution 3D mapping of cell positioning in the context of macroscale tissue. Given that the commonly used 2D immunofluorescence images or maximum intensity projections of 3D images significantly underestimate microglial cell motility [29, 35], such volumetric approach may be particularly relevant for evaluation of stereoscopic morphology of retinal microglial processes and their heterotypic interactions with other components of the neurovascular unit. The 3D reconstructed images enabled visualisation of extensively branched microglial cell processes that co-express two important components of the purinergic machinery, P2Y12R and CD39, and form direct contacts with exterior walls of CD39+ retinal blood vessels (Fig. 2A, Supplementary Fig. 3A, and Movie 1), as well as with neuronal cell bodies which either lack CD39 on their surface or merely express it at relatively low levels (Fig. 2B, Supplementary Fig. 3B, and Movies 2 and 3).
In situenzyme histochemistry and flow cytometric assays confirm cell type- and tissue-specific localization of ecto-nucleotidases in the mouse retina
In a different set of experiments, the activities of ecto-nucleotidases were measured in the mouse eye cryosections by using enzyme histochemistry assay . Additional staining of the samples with haematoxylin and eosin (H&E) (Fig. 3A) enabled the visualization of the main retinal layers and other ocular structures. The presence of dark light-absorbing melanin granules in the exterior retinal pigmented epithelium (RPE) partially interferes with lead nitrate-based enzyme histochemistry of the eye. Nevertheless, there were clear-cut differences in staining intensities between the samples incubated without (Fig. 3B) and with (Fig. 3C-E) exogenous nucleotides. High ATPase (Fig. 3C) and ADPase (Fig. 3D) activities were detected in the retinal vessels, OS of photoreceptor cells and neuronal bodies, while AMP-specific staining was mainly confined within the photoreceptor layer (Fig. 3E). High ATPase and ADPase (but not AMPase) activities were also detected in the stromal keratocytes and basal epithelial layer of the cornea (Fig. 3C-E). Notably, similar staining patterns were observed when eye cryosections were incubated with nucleotide substrates in the presence (Fig. 3C-E) and absence (data not shown) of the inhibitor of tissue-nonspecific alkaline phosphatase (TNAP) tetramisole. On the other hand, the use of the artificial chromogenic substrates of TNAP, BCIP and NBT, revealed the development of specific dark blue staining in the inner and outer plexiform layers of retina, as well as in the superficial corneal epithelial cell layer (Fig. 3F), which disappeared after pretreating the samples with tetramisole (data not shown). These data suggest that despite the selective expression of TNAP in certain eye structures, this broad substrate-specificity ectoenzyme is not implicated in the metabolism of ocular of ATP and other nucleotides. Collectively, in situ enzyme histochemistry, together with the multiplex imaging data described above (see Figs. 1 and 2), identified CD39 as the predominant ATP- and ADP-inactivating enzyme in the mouse eye which is expressed to varying degrees among vascular, immune, neural and stromal cells. The downstream step of hydrolysis of ATP/ADP-derived AMP into ADO is mediated through ecto-5’-nucleotidase/CD73 activity which is mainly localized in the photoreceptor layer. Flow cytometric analysis of isolated mouse retinal cells provided independent line of evidence for the presence of CD39 on CD45+/CD11b+/P2Y12R+ microglial cells (Fig. 3G) and CD45-/IB4+/CD31+ vascular endothelial cells (Fig. 3H). CD73 is also weakly expressed on microglial cells, but not in the blood vessels (Fig. 3G,H).
Single cell transcriptomic analysis of mouse and human retinal cells reveals relatively conserved purinergic signatures between the species
The expression profiles of genes encoding major purine-inactivating enzymes and ARs were also characterized at a single-cell resolution by using publicly available scRNAseq data of mouse retinal cells . Single-cell transcriptomic analysis demonstrated specific distribution of ectoenzymes in mouse vascular endothelial cells (Entpd1/CD39high, Alpl/TNAPhigh), perivascular cells (Entpd1/CD39low, Enpp1/ENPP1low), retinal microglia (Entpd1/CD39low), rod photoreceptors (Nt5e/CD73high, Alpl/TNAPlow), horizontal cells (Nt5e/CD73high), RGC (Entpd2/NTPDase2low), Müller glia and astrocytes (Entpd2/NTPDase2high) (Fig. 4A). In contrast to our multiplex imaging data showing the presence of CD39 immunoreactivity (Fig. 1B) and ATP/ADP-inactivating activity (Fig. 3C,D) in the OS of photoreceptor cells, transcriptomic approach did not reveal CD39-encoding gene in rod cells at mRNA level (Fig. 4A). Although the expression of Entpd1 on microglial cells was very low in this study, the use of another scRNAseq dataset of sorted Cx3cr1+ mouse retinal cells  revealed that Entpd1/CD39 is highly expressed on two major populations of P2ry12+ and Hmox1+ microglial cells, and additionally demonstrated the presence of other enzyme of the purine catabolic chain, ADA, on retinal Hmox1+ microglial cells and perivascular macrophages (Fig. 4B). While a detailed characterization of signal transduction pathways mediating biological effects of ADO lies beyond the scope of this study, we also analyzed the expression profiles of major AR subtypes. ARs are selectively expressed on various mouse retinal cells, including vascular endothelial (Adora2a/A2ARlow) and perivascular (Adora2a/A2ARhigh) cells, RGC (Adora1/A1Rhigh), Müller glia and astrocytes (Adora1/A1Rlow), P2ry12+ microglial cells and Rorb+ macrophages (Adora3/A3Rhigh) (Fig. 4A,B). Notably, data on highly selective expression of Nt5e/CD73 on the latter subset of Adora3+/Rorb+ macrophages suggest that ADO metabolism may be relevant in controlling adenosinergic signaling and function in this relatively small population of mouse retinal myeloid cells (Fig. 4B).
To further identify the similarities and differences in the purinergic signatures between rodent and human eyes, we utilized single cell transcriptomic atlas of the human retina . Ecto-nucleotidases and TNAP are selectively expressed on the human retinal endothelial cells (ENTPD1/CD39low, ALPL/TNAPhigh), microglial cells (ENTPD1/CD39high, NT5E/CD73low), rod terminals of photoreceptor cells (NT5E/CD73high, ALPL/TNAPlow), amacrine cells (ENPP1/ENPP1low), and Müller glia and astrocytes (ENTPD2/NTPDase2low), while the expression of ADO-inactivating enzyme ADA was maintained at very low or undetectable levels in all human retinal cells (Fig. 4C). These human transcriptomic data are consistent with our recent in situ enzyme histochemistry and immunofluorescence imaging data showing tissue-specific distribution of key ecto-nucleotidases (CD39, NTPDase2, CD73) and TNAP in the human sensory neuroretina and optic nerve head . Additional scRNAseq analysis of adenosinergic signaling pathways revealed that, similar to mouse retina, human retinal microglial cells and Müller glia and astrocytes express high levels of A3R (ADORA3) and A1R (ADORA1), respectively. However, unlike mouse blood vessels which express Adora2a/A2AR+, human retinal endothelial cells express another A2R subtype, ADORA2B/A2BR. Other human retinal cell subsets do not appear to express either AR subtype (Fig. 4C).
Overall, despite some species-specific variations, the expression profiles of key purinergic enzymes and AR subtypes appear to remain relatively conserved between the mouse and human eyes. These findings are summarized in Fig. 5 which schematically illustrates cell- and tissue-specific distribution of ecto-nucleotidases in the mammalian eye. In particular, data on selective compartmentalization of CD73 in both mouse and human photoreceptor cells provide a solid background for more thorough investigation of the role of this ectoenzyme in retinal function under various challenging and noxious conditions and further translation of these experimental data to clinic.
Pharmacological inhibition of ocular CD73 impairs retinal activity in dark-adapted mice exposed to bright light
Taking into account data on direct involvement of adenosinergic signaling in the modulation of light-evoked responses of retina [40–42], we hypothesized that pharmacological inhibition of the CD73-ADO axis may affect retinal function. Several novel CD73 inhibitors were designed and synthesized recently in our laboratories based on N6-benzyl-α,β-methylene ADP (PSB-12379) as a lead structure, which are characterized by exceptionally high selectivity, nanomolar inhibitory potency toward human, rat and mouse CD73, and high metabolic stability in human plasma and in rat liver microsomes . Studies with fluorescein-conjugated CD73 inhibitors additionally confirmed the utility of these compounds as fluorescent probes capable of binding directly to CD73 on various cells and tissues, including mouse CD73+ photoreceptor cells . Based on these observations, the most potent CD73 inhibitor, PSB-12489 (Fig. 6A) , was chosen as a suitable drug for further examination in our functional assays. Competitive TLC analysis of soluble CD73 activity confirmed the ability of PSB-12489 to inhibit the hydrolysis of [3H]AMP by human and mouse sera in a concentration-dependent manner with the IC50 values in the low nanomolar range, whereas the classical CD73 inhibitor adenosine 5′-(α,β-methylene)diphosphate (AMPCP) exerted inhibitory effects at ~ 100 times higher concentrations (Supplementary Fig. 4). This conclusion was independently ascertained by in situ enzyme histochemistry showing that treatment of mouse eye cryosections with increasing concentrations of PSB-12489 (0.1-1 µM), but not with equimolar concentrations of AMPCP, progressively reduced the intensity of AMP-specific staining in the photoreceptor layer (Fig. 6B).
Next, the role of CD73-generated ADO in the maintainance of retinal activity was assessed in vivo, as schematically illustrated in Fig. 6C. Electrophysiological analysis of the mouse retina was performed by recording fERG responses from dark-adapted (scotopic) eyes stimulated with increments of light intensity from 0.003 to 10 cd*s/m2 (Fig. 7A). Figure 7B shows representative electroretinograms recorded in mice from two experimental groups, which can be divided into the following components: the first a-wave that appears as a negative amplitude change, and the b-wave that appears as a large positive amplitude change right after the a-wave. The a-wave of the electroretinogram reflects the functional activity and integrity of the photoreceptors, whereas the b-wave originates in retinal cells that are post-synaptic to the photoreceptors, including inner retinal cells (bipolar and amacrine cells) and RGC [40, 44, 45]. Notably, C57BL/6N mice are known to be homozygous for the rd8 mutation in Crumbs homolog 1 (Crb1) gene, which may lead to severe retinal dysplasia in the inferior retina and other ocular abnormalities . These lesions appear as white to yellow flecks on fundus examination, and the phenotype is worsened by exposure of C57BL/6N mice to BL . Comparative analysis of fERG responses in C57BL/6N versus BALB/c mice also demonstated that C57BL/6N eyes are characterized by significant decreases in the b-wave amplitudes recorded at light intensities of 0.1 and 1 cd*s/m2, whereas the a-wave amplitudes remain comparable among the strains (Supplementary Fig. 5A).
Based on these findings, BALB/c mice were chosen as an appropriate model for studying the role of ADO metabolism in the retinal function. To achieve a sufficient inhibitory effect, PSB-12489 was administered locally into the vitreous cavity of dark-adapted mice at a relatively high dose (final concentration ~ 200 µM) followed by exposure of the treated mice to continuous illumination for 14 hours (Fig. 6C). Measurement of basal fERG values before treatment did not detect any differences in the a-wave and b-wave amplitudes in the study groups (Supplementary Fig. 5B). However, when fERG was repeated on day 7 post-treatment, relatively moderate but significant decreases in the b-wave amplitudes were found in the PSB-12489-treated eyes (group G2), when compared to vehicle-treated (G3) and non-treated control (G1) groups (Fig. 7C). These differences became even more substantial after exposing the PSB-12489-treated animals to continuous BL (group G5). These mice were characterized by ~ 40–50% decrease in the b-wave amplitudes at all stimulus levels of the light intensity tested and also showed a decrease in the a-wave amplitude recorded at high light intensities (1–10 cd*s/m2), when compared to vehicle-treated (G6) and non-treated (G4) groups exposed to BL (Fig. 7D). Notably, in contrast to the conventional experimental model of BL-induced retinal damage induced by extending the period of dark adaptation to 24 hours and characterized by markedly impaired scotopic responses (our unpublished observations), the combination of dark adaptation and light illumination parameters used in this work did not by itself cause any adverse effects on retinal electrical activity (Supplementary Fig. 5C).
Exposure of dark-adapted eyes to CD73 inhibitor and BL caused a loss of RGC with no changes in total retinal thickness.
The thickness of the retina was determined in live animals immediately after fERG recording by using high-resolution spectral domain optical coherence tomography (OCT). Photoreceptor layer thickness was measured in superior temporal area of the retina, which is the most sensitive to the retinal damage. No significant changes in total retinal thickness were observed between the groups studied (Fig. 7E). Photoreceptor CD73 activity and the integrity of the treated retina were also interrogated at the histological level. Given the uneven distribution of CD73 in the mouse photoreceptor layer, AMP-specific staining intensities were determined in three different regions, including highly CD73-positive OS of photoreceptor cells, as well as ONL and OPL characterized by intermediate enzyme expression (Fig. 8A, left panel). Quantitative analysis did not detect any down-regulation of AMPase activity in the eyes receiving PSB-12489 (groups G2 and G5; Fig. 8A). These data suggest that, in spite of high inhibitory potency of PSB-12489 toward photoreceptor CD73 (Fig. 6B), a single intravitreal injection of PSB-12489 presumably inhibits ocular ADO production via a temporary and reversible mechanism, which restored to control levels on day 7 post-treatment. The numbers of nuclei in the retinal layers were quantified to provide a further assessment of cell survival in the treated eyes. Mice receiving PSB-12489 and exposed to BL (G5) were characterized by a loss of RGC in the innermost GCL, with no change in the total number of photoreceptor cells present in the inner and outer segments of the retina (Fig. 8B).
Further studies would be required to elucidate the exact mechanism(s) underlying the revealed inhibitory effects of PSB-12489 on retinal function. Taking into account the complexity of purine homeostasis in the mammalian eye (see Fig. 5), and the important role of adenosinergic signaling in retinal functioning during transition from darkness to light [11, 12], it is reasonable to conclude that inhibition of the CD73-ADO axis in dark-adapted mice could shift the entire balance between ocular ATP and ADO levels. Potential consequences of blocking this metabolic chain are highlighted in Fig. 9 and may particularly include the impaired neuroprotective and hyperemic responses mediated via AR-dependent and/or intrinsic receptor-independent mechanisms, as well as the simultaneous accumulation of pro-inflammatory and cytotoxic ATP in the retinal environment.